CN108765465A - A kind of unsupervised SAR image change detection - Google Patents

A kind of unsupervised SAR image change detection Download PDF

Info

Publication number
CN108765465A
CN108765465A CN201810549675.XA CN201810549675A CN108765465A CN 108765465 A CN108765465 A CN 108765465A CN 201810549675 A CN201810549675 A CN 201810549675A CN 108765465 A CN108765465 A CN 108765465A
Authority
CN
China
Prior art keywords
sample
pixel
sar image
obtains
training sample
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201810549675.XA
Other languages
Chinese (zh)
Other versions
CN108765465B (en
Inventor
李明
李梦柯
吴艳
张鹏
刘慧敏
柴磊
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xidian University
Original Assignee
Xidian University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Xidian University filed Critical Xidian University
Priority to CN201810549675.XA priority Critical patent/CN108765465B/en
Publication of CN108765465A publication Critical patent/CN108765465A/en
Application granted granted Critical
Publication of CN108765465B publication Critical patent/CN108765465B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
    • G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10032—Satellite or aerial image; Remote sensing
    • G06T2207/10044—Radar image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Multimedia (AREA)
  • Probability & Statistics with Applications (AREA)
  • Image Analysis (AREA)

Abstract

The present invention proposes a kind of unsupervised SAR image change detection, the technical problem relatively low for solving detection accuracy and computational efficiency existing in the prior art.Realize that step is：To the two phase SAR images filtering after registration；Generate initial difference figure；Conspicuousness detection is carried out to initial difference figure and obtains significant difference figure；Significant difference figure presort using Fuzzy C-Means Cluster Algorithm and obtains candidate training sample and uncertain sample；Equalization processing is done to candidate candidate training sample set and obtains training sample；Decent uncertain sample characteristics of seeking peace of training are extracted with PCA filters；With training sample feature Training Support Vector Machines, the support vector machines completed with training obtains variation testing result to not knowing sample classification.The present invention, which can improve, to be calculated detection accuracy while reducing operation time, be can be used for assessing disaster and is predicted disaster development trend, target acquisition and land cover pattern and utilize monitoring etc..

Description

A kind of unsupervised SAR image change detection
Technical field
The invention belongs to technical field of image processing, are related to a kind of image change detection method, and in particular to a kind of improvement The unsupervised SAR image change detections of PCANet, can be used for assessing disaster and predict disaster development trend, target acquisition with And land cover pattern and utilization monitoring etc..
Background technology
Image Change Detection refer to the same earth surface area obtained using multidate remote sensing image and other auxiliary datas come Determine and analysis earth's surface variation, be update geodata key technology, assessment disaster and prediction disaster development trend and It land cover pattern and is played an important role using in terms of monitoring.It is to have with the microwave remote sensing that synthetic aperture radar (SAR) is representative Source microwave imaging sensor.Compared with the microwave remote sensings such as optical remote sensing, SAR imaging techniques not only have variable side view angle, and Detailed geography information can be obtained in the case that weatherproof, therefore have in change detection techniques and widely answer With.According to the needs of handmarking's sample, can be divided into image change detection method has supervised, Semi-supervised and unsupervised formula Three kinds.There are supervision and Semi-supervised method that there is very strong dependence to marker samples, in SAR image processing, marker samples It is not easy to obtain.Unsupervised approaches do not need marker samples and manual intervention, therefore in SAR image change detection more often With.
Unsupervised approaches need that all samples are first divided into different classifications automatically, then select in the application of variation detection Go out some representative samples as training sample, is used for the training of grader.During this, the selection of training sample It can influence to change accuracy in detection.
Since deep learning method has good behaviour, SAR image variation detection in image classification and field of face identification In grader can utilize deep learning method.Gao F. et al. in 2016《Geoscience Remote Sensing Letters》On deliver an entitled " Automatic change detection in synthetic aperture radar The article of images based on PCANet " is changed detection using PCANet to SAR image for the first time.This method is used first Gabor filter extracts Gabor characteristic to original image, with fuzzy C-means clustering method according to image Gabor characteristic by pixel It is divided into variation class, non-changing class and uncertain class in advance, then the pixel for belonging to uncertain class is carried out using PCANet models Classification.This method can obtain accurately variation testing result effectively to not knowing pixel classifications.However there are still it is following not Foot：(1) because SAR image is influenced by coherent speckle noise, the training sample that this method is chosen from original image is unreliable, and feature carries This method does not account for the unbalanced problem of training sample when taking, and causes to the imperfect of Different categories of samples feature extraction, in addition should Method uses mode pixel-based, does not account for neighborhood of pixels information, partial pixel mistake point can be caused, to make variation detect Accuracy reduces；(2) this method selects training sample in all pixels, causes processing time longer.
Invention content
It is an object of the invention to overcome the shortcomings of above-mentioned prior art, it is proposed that a kind of unsupervised SAR image variation inspection Survey method, for solving detection accuracy existing in the prior art and less efficient technical problem.
The present invention technical thought be：Place is filtered to two phase SAR images after registration with three-dimensional bits matched filtering Reason；Generate initial difference figure；Conspicuousness detection is carried out to initial difference figure and obtains significant difference figure；It is poly- using fuzzy C-mean algorithm Class algorithm, which to significant difference figure presort, obtains candidate training sample and uncertain sample；Method is chosen using consistent Equalization processing is done to candidate candidate training sample set and obtains training sample；With PCA filters extract training pixel characteristic and Uncertain pixel characteristic；With training pixel characteristic Training Support Vector Machines, the support vector machines completed with training is not to knowing picture Element classification obtains variation testing result.
According to above-mentioned technical thought, realize that the technical solution that the object of the invention is taken includes the following steps：
(1) SAR image to be detected to two is filtered：
The two width SAR images to shooting in same place different moments are registrated, and are carried out to the SAR image after registration Filtering, obtains filtered SAR image im1And im2；
(2) filtered SAR image im is calculated1And im2Initial difference figure D1；
(3) to D1Conspicuousness detection is carried out, and binaryzation is carried out to testing result：
To D1Conspicuousness detection is carried out, D is obtained1Notable figure Ds, and to DsBinaryzation is carried out, binaryzation notable figure is obtained Ds'；
(4) D is obtained1Significant difference figure D2：
Utilize binaryzation notable figure Ds', extraction initial difference figure D1In conspicuousness part, obtain D1Significant difference Scheme D2：
D2=DotM (Ds', D1)
Wherein, DotM (Ds', D1) indicate Ds' and D1The gray value of same position pixel is multiplied；
(5) to D1Significant difference figure D2In pixel presort：
To significant difference figure D2In pixel clustered, obtain positive sample collection, negative sample collection and uncertain sample set, And positive sample collection and negative sample collection are merged into candidate training sample set；
(6) candidate training sample set is equalized：
The positive sample collection and negative sample collection concentrated to candidate training sample carry out equalization processing, the training for being equalized Sample set, and from 30% sample is wherein randomly selected as training sample set；
(7) training sample feature and uncertain sample characteristics are extracted：
(7a) is to im1There is overlapping to divide pixel-by-pixel, obtains the multiple images block for s × s by sizeThe collection of composition Close I1, while to im2There is overlapping to divide pixel-by-pixel, obtains the multiple images block that size is s × sThe set I of composition2：
Wherein, s is the odd number more than or equal to 1, and pos indicates that image block serial number, pos=1,2 ..., T, T indicate SAR image im1Or im2The number of middle image block；
Training sample set and uncertain sample set are merged into sample set by (7b), and from I1Middle selection and sample in sample set Position xaCorresponding image blockSimultaneously from I2Middle selection and sample position x in sample setaCorresponding image blockA tables This serial number of sample, and a=1,2 ..., NE, NEFor the sum of sample in sample set；
(7c) is rightThere is overlapping to divide pixel-by-pixel, obtains multiple neighborhood block P for m × m by sizei 1, while it is right There is overlapping to divide pixel-by-pixel, obtains multiple neighborhood block P that size is m × mi 2, and by neighborhood block Pi 1And Pi 2The neighbour being merged into Domain block PiIt is combined to set PtempIn：
Wherein, m is the odd number for being less than or equal to s more than or equal to 1, and i indicates that neighborhood block serial number, and i=1,2 ..., n, n are indicated Neighborhood block number；
(7d) is to PtempIn neighborhood block PiCarry out vectorization again after carrying out mean value, obtain include byThe vector matrix P of composition：
(7e) calculates PPTFeature vector, and by PPTPreceding N1A feature vector is as first layer PCA filters Wl 1：
Wl 1=mat (ql(PPT))∈R2m×mL=1,2 ..., N1
Wherein, N1For positive integer, ql(PPT) indicate PPTFirst of feature vector, mat (v) indicate vector Mapping becomes a matrix W ∈ R2m×m；
(7f) passes through first layer PCA filters Wl 1Calculate sample PiCharacteristic information Pi l：
Pi l=Pi*Wl 1
Wherein, * indicates Three dimensional convolution operation；
(7g) is to first layer filter Wl 1The characteristic information P of outputi lCarry out vectorization again after mean value, obtain include byThe vector matrix Q of composition：
(7h) calculates QQTFeature vector, and by QQTPreceding N2A feature vector is as second layer PCA filters
Wherein, qk(QQT) indicate QQTK-th of feature vector, k=1,2 ..., N2, N2For positive integer, mat (v) is indicated VectorMapping becomes a matrix W ∈ R2m×m；
(7i) passes through second layer PCA filtersCalculate characteristic information Pi lQuadratic character information
(7j) is using He Wei Sadens jump function to characteristic information matrixBinaryzation is carried out, and The binaryzation information matrix that will be obtainedThe numerical value for being converted to each position isMatrix O after, meter Calculate PiFeature fi, and willAs training sample feature, As uncertain sample characteristics, wherein：
Wherein,Expression pairStatistics with histogram, NTIndicate that training sample concentrates the number of sample；
(8) the variation testing result of SAR image is obtained：
Using training sample feature as the input of support vector machines, trained support vector machines is obtained, and will not know Input of the sample characteristics as trained support vector machines obtains the variation testing result figure of SAR image.
Compared with prior art, the present invention haing the following advantages：
(1) present invention uses during obtaining training sample using the gray value of neighborhood block as cluster feature, avoids existing There is technology to cause the insecure defect of training sample by cluster feature of the Gabor characteristic of original image；Simultaneously in feature extraction On the one hand Cheng Zhong equalizes candidate training sample set using consistent selection method, can effectively eliminate imbalanced training sets and ask Topic, avoids the incomplete defect of Different categories of samples feature extraction that the prior art is directly extracted caused by sample extraction feature, On the other hand it by the way of based on neighborhood block, avoids the prior art and uses partial pixel caused by mode pixel-based wrong The defect divided, the experimental results showed that the present invention can effectively improve variation detection accuracy.
(2) present invention extracts salient region with Context-aware conspicuousness detection methods, is selected in salient region Training sample and uncertain sample are selected, avoids that the prior art chooses training sample from entire image and uncertain sample causes The big defect of sample size, the experimental result surface present invention can effectively reduce operation time, improve the effect of Image Change Detection Rate.
Description of the drawings
Fig. 1 is the implementation flow chart of the present invention；
Fig. 2 is the variation testing result pair that the present invention and the prior art are applied to two phase Bern data Real SAR images Than figure；
Fig. 3 is the variation detection that the present invention and the prior art are applied to two phase the Yellow River estuary data Real SAR images Comparative result figure；
Fig. 4 is the variation testing result that the present invention and the prior art are applied to two phase Ottawa data Real SAR images Comparison diagram.
Specific implementation mode
In the following with reference to the drawings and specific embodiments, present invention is further described in detail：
A kind of unsupervised SAR image change detections of referring to Fig.1, include the following steps：
Step 1) is filtered two SAR images to be detected：
Image imo 1For the SAR image that the moment 1 shoots, image imo 2For the SAR image that the moment 2 shoots, imo 1And imo 2Shooting With same place, to imo 1And imo 2It is registrated, using three Block- matching filtering methods, to the SAR image after registration with carrying out Filtering removes the coherent speckle noise in image and keeps image boundary information, obtains filtered image im1And im2：
im1={ im1(i, j) | 1≤i≤H, 1≤j≤W },
im2={ im2(i, j) | 1≤i≤H, 1≤j≤W },
H indicates that the height of acquired SAR image, W indicate that the width of acquired SAR image, (i, j) indicate in image The position of pixel.
Step 2) obtains filtered SAR image im1And im2Initial difference figure D1：
Using logarithm ratio algorithm, filtered SAR image im is calculated1And im2Initial difference figure D1, calculation formula is：
Step 3) is to D1Conspicuousness detection is carried out, and binaryzation is carried out to testing result：
Using Context-aware conspicuousness detection methods to D1Conspicuousness detection is carried out, D is obtained1Notable figure Ds, and To DsBinaryzation is carried out, binaryzation notable figure D is obtaineds', calculation formula is：
Wherein, p (x, y) indicates notable figure DsPosition is the gray value of the pixel of (x, y), and τ expressions are calculated by big Tianjin Method calculates the threshold value obtained, and 1 indicates conspicuousness pixel, and 0 indicates non-limiting pixel.
Step 4) extracts D1In conspicuousness part：
Utilize binaryzation notable figure Ds', to initial difference figure D1In conspicuousness part extract, it is poor to obtain conspicuousness Different figure D2：
D2=DotM (Ds', D1)
Wherein, DotM (Ds', D1) indicate Ds' and D1The gray value of same position pixel is multiplied.
Step 5) is to significant difference figure D2In pixel presort：
Step 5a) with significant difference figure D2In pixel xiCentered on, by D2It is divided into the neighborhood block that size is m × mThe value of m is positive integer, and m is set as 3 in the embodiment of the present invention, can effectively inhibit noise and image boundary is kept to believe Breath；
Step 5b) Fuzzy C-Means Cluster Algorithm is used, withThe gray value of middle all pixels is cluster feature, to D2In Pixel clustered, obtain n class pixels, n is the integer more than or equal to 3, in order to ensure sample authenticity, the embodiment of the present invention Middle n is set as 5, then obtains 5 class pixels, this five class is arranged in descending order according to the size of cluster centre modulus value and obtains c1, c2,...,c5；
Step 5c) by c1In each pixel be labeled as positive sample, obtain positive sample collection cP, while by c5In each picture Element label is to obtain negative sample collection cP, by c2,c3And c4In pixel be labeled as uncertain sample and obtain uncertain sample This collection cU。
Step 5d) by cPAnd cNMerge into candidate training sample set.
Step 6) equalizes candidate training sample set：
Step 6a) assume positive sample collection cPMiddle number of samples is nP, negative sample collection cNMiddle number of samples is nN, and nP< nN, Duplication ratio is N：
Wherein, [] indicates bracket function；
Step 6b) by cPIn sample replicate N part, form set cP', to cP' and cNMiddle sample merges, and from conjunction And 30% sample is randomly selected in result as training sample.
Step 7) extracts sample characteristics with PCA filters：
Step 7a) to im1There is overlapping to divide pixel-by-pixel, obtains the multiple images block for s × s by sizeComposition Set I1, while to im2There is overlapping to divide pixel-by-pixel, obtains the multiple images block that size is s × sThe set of composition I2：
Wherein, s is the odd number more than or equal to 1, and pos indicates that image block serial number, pos=1,2 ..., T, T indicate SAR image im1Or im2The number of middle image block, T=WH equal with number of pixels in image, the value that s in example is applied in the invention of this reality are 13；
Step 7b) training sample set and uncertain sample set merged into sample set, and from I1In middle selection and sample set Sample position xaCorresponding image blockSimultaneously from I2Middle selection and sample position x in sample setaCorresponding image blockA indicates sample serial number, and a=1,2 ..., NE, NEFor the sum of sample in sample set, wherein positive sample number is NT, no Determine that number of samples is N in sampleU, then have NE=NT+NU；
Step 7c) it is rightThere is overlapping to divide pixel-by-pixel, obtains multiple neighborhood block P for m × m by sizei 1, simultaneously It is rightThere is overlapping to divide pixel-by-pixel, obtains multiple neighborhood block P that size is m × mi 2, and by neighborhood block Pi 1And Pi 2It is merged into Neighborhood block PiIt is combined to set PtempIn：
Wherein, m is the odd number for being less than or equal to s more than or equal to 1, and i indicates that neighborhood block serial number, and i=1,2 ..., n, n are indicated Neighborhood block number, and n=ssNE, the value of m is 3 in the embodiment of the present invention；
Step 7d) to PtempIn neighborhood block PiCarry out vectorization again after carrying out mean value, obtain include byThe vector matrix P of composition：
Step 7e) calculate PPTFeature vector, and by PPTPreceding N1A feature vector is as first layer PCA filters Wl 1：
Wl 1=mat (ql(PPT))∈R2m×mL=1,2 ..., N1
Wherein, N1For positive integer, ql(PPT) indicate PPTFirst of feature vector, mat (v) indicate vector Mapping becomes a matrix W ∈ R2m×m, N in the present embodiment1Value be 8, indicate upper and lower, left and right, upper left, lower-left, upper right, The direction of bottom right 8；
Step 7f) pass through first layer PCA filters Wl 1Calculate sample PiCharacteristic information Pi l：
Pi l=Pi*Wl 1
Wherein, * indicates Three dimensional convolution operation；
Step 7g) to first layer filter Wl 1The characteristic information P of outputi lVectorization again after mean value is carried out, including ByThe vector matrix Q of composition：
Step 7h) calculate QQTFeature vector, and by QQTPreceding N2A feature vector is as second layer PCA filters
Wherein, qk(QQT) indicate QQTK-th of feature vector, k=1,2 ..., N2, N2For positive integer, mat (v) is indicated VectorMapping becomes a matrix W ∈ R2m×m, N in the present embodiment2Value be 8, indicate upper and lower, left and right, a left side Upper, lower-left, upper right, the direction of bottom right 8；
Step 7i) pass through second layer PCA filtersCalculate characteristic information Pi lQuadratic character information
Step 7j) use He Wei Sadens jump function to characteristic information matrixCarry out two-value Change, calculation formula is：
The binaryzation information matrix that will be obtainedThe numerical value for being converted to each position isMatrix O Afterwards, P is calculatediFeature fi, and willAs training sample feature, As uncertain sample characteristics, it is ssN not know sample characteristics numberU, wherein：
Wherein,Expression pairStatistics with histogram；
Step 8) Training Support Vector Machines and to do not know sample classification：
Using training sample feature as the input of support vector machines, trained support vector machines is obtained, and will not know Input of the sample characteristics as trained support vector machines obtains the variation testing result of SAR image.
Below in conjunction with emulation experiment, further details of analysis is made to the technique effect of the present invention.
1, experiment condition
Emulation is in Core i5-6500CPU, dominant frequency 3.2GHz, the hardware environment and MATLAB of memory 4GB below It is carried out under the software environment of 2015b.
2, experiment content
Experiment one：Two phase of detection method pair is changed to SAR image using PCANet with existing using the present invention Bern data Real SAR images are changed detection, and the results are shown in Figure 2.
Experiment two：Two phase of detection method pair is changed to SAR image using PCANet with existing using the present invention The Yellow River estuary data Real SAR images are changed detection, and the results are shown in Figure 3.
Experiment three：Two phase of detection method pair is changed to SAR image using PCANet with existing using the present invention Ottawa data Real SAR images are changed detection, and the results are shown in Figure 4.
3, experimental result and analysis
With reference to Fig. 2, wherein Fig. 2 (a) is the first moment Bern data set SAR image, and Fig. 2 (b) is the second moment Bern number According to SAR image, Fig. 2 (c) is variation testing result reference chart, and Fig. 2 (d) is the variation testing result of the prior art, and Fig. 2 (e) is The variation testing result of the present invention.
It is the first moment the Yellow River estuary data SAR image with reference to Fig. 3, wherein Fig. 3 (a), Fig. 3 (b) is the second moment Huang River estuary data SAR image, Fig. 3 (c) are variation testing result reference charts, and Fig. 3 (d) is the variation detection knot of the prior art Fruit, Fig. 3 (e) are the variation testing results of the present invention.
With reference to Fig. 4, wherein Fig. 4 (a) is the first moment Ottawa data SAR image, and Fig. 4 (b) is the second moment Ottawa Data SAR image, Fig. 4 (c) are variation testing result reference charts, and Fig. 4 (d) is the variation testing result of the prior art, Fig. 4 (e) It is the variation testing result of the present invention.
Change testing result quality for verification, the present invention chooses overall accuracy, Kappa coefficients and executes the time as performance Index parameter evaluates the detection accuracy and computational efficiency of the present invention.
The experimental result is as shown in Table 1 and Table 2, and table 1 provides the comparison of the prior art and detection accuracy of the present invention, table 2 Provide the comparison of the prior art and computational efficiency of the present invention.
1 prior art of table is compared with accuracy of detection of the present invention
2 prior art of table is compared with efficiency of the present invention
It can see by Fig. 2, Fig. 3 and Fig. 4, the variation testing result that the present invention obtains provides result more with reference chart It is close.By table 1, it can be seen that, the present invention is better than the prior art in accuracy of detection.This is because obtaining training sample process In, the present invention is used using the gray value of neighborhood block as cluster feature, keeps training sample more reliable, and sample is eliminated with consistent selection method This is unbalanced, keeps the positive negative training sample feature of extraction more complete, and pixel neighbour is fully considered using block-based processing mode Domain information simultaneously effectively inhibits coherent speckle noise, so that the feature of extraction is preferably reflected variation sample and does not change sample Difference, therefore the variation accuracy in detection of the present invention is apparently higher than the prior art.
It can see by table 2, computational efficiency higher of the present invention.This is because the present invention chooses training from salient region Sample and uncertain sample effectively reduce sample size, therefore computational efficiency of the present invention is higher than the prior art.

Claims (6)

1. a kind of unsupervised SAR image change detection, which is characterized in that include the following steps：
(1) SAR image to be detected to two is filtered：
The two width SAR images to shooting in same place different moments are registrated, and are filtered to the SAR image after registration Wave obtains filtered SAR image im1And im2；
(2) filtered SAR image im is calculated1And im2Initial difference figure D1；
(3) to D1Conspicuousness detection is carried out, and binaryzation is carried out to testing result：
To D1Conspicuousness detection is carried out, D is obtained1Notable figure Ds, and to DsBinaryzation is carried out, binaryzation notable figure D is obtaineds'；
(4) D is obtained1Significant difference figure D2：
Utilize binaryzation notable figure Ds', extraction initial difference figure D1In conspicuousness part, obtain D1Significant difference figure D2：
D2=DotM (Ds', D1)
Wherein, DotM (Ds', D1) indicate Ds' and D1The gray value of same position pixel is multiplied；
(5) to D1Significant difference figure D2In pixel presort：
To significant difference figure D2In pixel clustered, obtain positive sample collection, negative sample collection and uncertain sample set, and will Positive sample collection and negative sample collection merge into candidate training sample set；
(6) candidate training sample set is equalized：
The positive sample collection and negative sample collection concentrated to candidate training sample carry out equalization processing, the training sample for being equalized Collection, and from 30% sample is wherein randomly selected as training sample set；
(7) training sample feature and uncertain sample characteristics are extracted：
(7a) is to im1There is overlapping to divide pixel-by-pixel, obtains the multiple images block for s × s by sizeThe set I of composition1, Simultaneously to im2There is overlapping to divide pixel-by-pixel, obtains the multiple images block that size is s × sThe set I of composition2：
Wherein, s is the odd number more than or equal to 1, and pos indicates that image block serial number, pos=1,2 ..., T, T indicate SAR image im1Or im2The number of middle image block；
Training sample set and uncertain sample set are merged into sample set by (7b), and from I1Middle selection and sample position in sample set xaCorresponding image blockSimultaneously from I2Middle selection and sample position x in sample setaCorresponding image blockA indicates sample This serial number, and a=1,2 ..., NE, NEFor the sum of sample in sample set；
(7c) is rightThere is overlapping to divide pixel-by-pixel, obtains multiple neighborhood block P for m × m by sizei 1, while it is rightIt carries out There is overlapping to divide pixel-by-pixel, obtains multiple neighborhood block P that size is m × mi 2, and by neighborhood block Pi 1And Pi 2The neighborhood block being merged into PiIt is combined to set PtempIn：
Wherein, m is the odd number for being less than or equal to s more than or equal to 1, and i indicates that neighborhood block serial number, and i=1,2 ..., n, n indicate neighborhood Block number；
(7d) is to PtempIn neighborhood block PiCarry out vectorization again after carrying out mean value, obtain include byComposition Vector matrix P：
(7e) calculates PPTFeature vector, and by PPTPreceding N1A feature vector is as first layer PCA filters Wl 1：
Wl 1=mat (ql(PPT))∈R2m×mL=1,2 ..., N1
Wherein, N1For positive integer, ql(PPT) indicate PPTFirst of feature vector, mat (v) indicate vectorMapping As a matrix W ∈ R2m×m；
(7f) passes through first layer PCA filters Wl 1Calculate sample PiCharacteristic information Pi l：
Pi l=Pi*Wl 1
Wherein, * indicates Three dimensional convolution operation；
(7g) is to first layer filter Wl 1The characteristic information P of outputi lCarry out vectorization again after mean value, obtain include byThe vector matrix Q of composition：
(7h) calculates QQTFeature vector, and by QQTPreceding N2A feature vector is as second layer PCA filters
Wherein, qk(QQT) indicate QQTK-th of feature vector, k=1,2 ..., N2, N2For positive integer, mat (v) indicate to AmountMapping becomes a matrix W ∈ R2m×m；
(7i) passes through second layer PCA filtersCalculate characteristic information Pi lQuadratic character information
(7j) is using He Wei Sadens jump function to characteristic information matrixProgress binaryzation, and will The binaryzation information matrix arrivedThe numerical value for being converted to each position isMatrix O after, calculate Pi Feature fi, and willAs training sample feature,Make Not know sample characteristics, wherein：
Wherein,Expression pairStatistics with histogram, NTIndicate that training sample concentrates the number of sample；
(8) the variation testing result of SAR image is obtained：
Using training sample feature as the input of support vector machines, trained support vector machines is obtained, and by uncertain sample Input of the feature as trained support vector machines obtains the variation testing result figure of SAR image.
2. a kind of unsupervised SAR image change detection according to claim 1, which is characterized in that institute in step (1) That states is filtered the SAR image after registration, using three-dimensional bits matched filtering method.
3. a kind of unsupervised SAR image change detection according to claim 1, which is characterized in that institute in step (2) The filtered SAR image im of calculating stated1And im2Initial difference figure D1, using logarithm ratio algorithm, calculation formula is：
4. a kind of unsupervised SAR image change detection according to claim 1, which is characterized in that institute in step (3) The binaryzation notable figure D stateds', calculation formula is：
Wherein, p (x, y) indicates notable figure DsPosition is the gray value of the pixel of (x, y), DsIt indicates to utilize Context- Aware conspicuousness detection methods are to D1The detected notable figure of conspicuousness is carried out, τ indicates to calculate acquisition by Otsu algorithm Threshold value, 1 indicates conspicuousness pixel, and 0 indicates non-limiting pixel.
5. a kind of unsupervised SAR image change detection according to claim 1, which is characterized in that institute in step (5) State to significant difference figure D2In pixel clustered, realize step be：
(5a) is with significant difference figure D2In pixel xbCentered on, by D2It is divided into the neighborhood block that size is size × sizeThe value of size is positive integer；
(5b) uses Fuzzy C-Means Cluster Algorithm, withThe gray value of middle all pixels is cluster feature, to D2In pixel into Row cluster, obtains n class pixels, and n is the integer more than or equal to 3；
Each pixel in cluster centre modulus value maximum kind in n class pixels is labeled as positive sample by (5c), obtains positive sample collection cP, Each pixel in cluster centre modulus value infima species is labeled as negative sample simultaneously, obtains negative sample collection cN, and will be in remaining class Each pixel be labeled as uncertain sample, obtain uncertain sample set cU。
6. a kind of unsupervised SAR image change detection according to claim 1, which is characterized in that institute in step (6) The positive sample collection c that candidate training sample is concentrated statedPWith negative sample collection cNEqualization processing is carried out, chooses method using consistent, in fact Now step is：
(6a) assumes positive sample collection cPMiddle number of samples is nP, negative sample collection cNMiddle number of samples is nN, and nP< nN, replicate ratio For N：
Wherein, [] indicates bracket function；
(6b) is by cPIn sample replicate N part, form set cP', to cP' and cNMiddle sample merges, and from amalgamation result 30% sample is randomly selected as training sample.
CN201810549675.XA 2018-05-31 2018-05-31 Unsupervised SAR image change detection method Active CN108765465B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810549675.XA CN108765465B (en) 2018-05-31 2018-05-31 Unsupervised SAR image change detection method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810549675.XA CN108765465B (en) 2018-05-31 2018-05-31 Unsupervised SAR image change detection method

Publications (2)

Publication Number Publication Date
CN108765465A true CN108765465A (en) 2018-11-06
CN108765465B CN108765465B (en) 2020-07-10

Family

ID=64001245

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810549675.XA Active CN108765465B (en) 2018-05-31 2018-05-31 Unsupervised SAR image change detection method

Country Status (1)

Country Link
CN (1) CN108765465B (en)

Cited By (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109598291A (en) * 2018-11-23 2019-04-09 安徽大学 A kind of collaboration well-marked target detection method of the RGBD figure based on PSO
CN109871875A (en) * 2019-01-21 2019-06-11 大连理工大学 A kind of building change detecting method based on deep learning
CN109934281A (en) * 2019-03-08 2019-06-25 电子科技大学 A kind of unsupervised training method of two sorter networks
CN109961028A (en) * 2019-03-15 2019-07-02 西安电子科技大学 SAR detection method based on three-dimensional Block- matching and full condition of contact random field
CN110176027A (en) * 2019-05-27 2019-08-27 腾讯科技（深圳）有限公司 Video target tracking method, device, equipment and storage medium
CN110781922A (en) * 2019-09-27 2020-02-11 北京淇瑀信息科技有限公司 Sample data generation method and device for machine learning model and electronic equipment
CN111080678A (en) * 2019-12-31 2020-04-28 重庆大学 Multi-temporal SAR image change detection method based on deep learning
CN111275680A (en) * 2020-01-18 2020-06-12 中国海洋大学 SAR image change detection method based on Gabor convolution network
CN112288744A (en) * 2020-08-24 2021-01-29 西安电子科技大学 SAR image change detection method based on integer reasoning quantification CNN
CN113033510A (en) * 2021-05-21 2021-06-25 浙江大华技术股份有限公司 Training and detecting method, device and storage medium for image change detection model
CN113569950A (en) * 2021-07-28 2021-10-29 大唐环境产业集团股份有限公司 Power station equipment fault monitoring model generation method, system and device
CN116206215A (en) * 2023-03-17 2023-06-02 银河航天(北京)网络技术有限公司 Forest land state monitoring method, forest land state monitoring device and storage medium
CN116990298A (en) * 2023-09-28 2023-11-03 南通中奥车用新材料有限公司 Finished product quality evaluation system for artificial leather production equipment

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20030190075A1 (en) * 2001-10-31 2003-10-09 Infowrap, Inc. Method for illumination independent change detection in a pair of registered gray images
US20120328161A1 (en) * 2011-06-22 2012-12-27 Palenychka Roman Method and multi-scale attention system for spatiotemporal change determination and object detection
CN103456018A (en) * 2013-09-08 2013-12-18 西安电子科技大学 Remote sensing image change detection method based on fusion and PCA kernel fuzzy clustering

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20030190075A1 (en) * 2001-10-31 2003-10-09 Infowrap, Inc. Method for illumination independent change detection in a pair of registered gray images
US20120328161A1 (en) * 2011-06-22 2012-12-27 Palenychka Roman Method and multi-scale attention system for spatiotemporal change determination and object detection
CN103456018A (en) * 2013-09-08 2013-12-18 西安电子科技大学 Remote sensing image change detection method based on fusion and PCA kernel fuzzy clustering

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
GAO ET AL: "《Automatic Change Detection in Synthetic Aperture Radar Images Based on》", 《IEEE GEOSCIENCE AND REMOTE SENSING LETTERS 》 *
慕彩红等: "《基于小波融合和PCA-核模糊聚类的遥感图像变化检测》", 《电子学报》 *

Cited By (20)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109598291A (en) * 2018-11-23 2019-04-09 安徽大学 A kind of collaboration well-marked target detection method of the RGBD figure based on PSO
CN109871875B (en) * 2019-01-21 2021-01-19 大连理工大学 Building change detection method based on deep learning
CN109871875A (en) * 2019-01-21 2019-06-11 大连理工大学 A kind of building change detecting method based on deep learning
CN109934281A (en) * 2019-03-08 2019-06-25 电子科技大学 A kind of unsupervised training method of two sorter networks
CN109961028A (en) * 2019-03-15 2019-07-02 西安电子科技大学 SAR detection method based on three-dimensional Block- matching and full condition of contact random field
CN110176027A (en) * 2019-05-27 2019-08-27 腾讯科技（深圳）有限公司 Video target tracking method, device, equipment and storage medium
CN110176027B (en) * 2019-05-27 2023-03-14 腾讯科技（深圳）有限公司 Video target tracking method, device, equipment and storage medium
CN110781922A (en) * 2019-09-27 2020-02-11 北京淇瑀信息科技有限公司 Sample data generation method and device for machine learning model and electronic equipment
CN111080678A (en) * 2019-12-31 2020-04-28 重庆大学 Multi-temporal SAR image change detection method based on deep learning
CN111275680A (en) * 2020-01-18 2020-06-12 中国海洋大学 SAR image change detection method based on Gabor convolution network
CN111275680B (en) * 2020-01-18 2023-05-26 中国海洋大学 SAR image change detection method based on Gabor convolution network
CN112288744A (en) * 2020-08-24 2021-01-29 西安电子科技大学 SAR image change detection method based on integer reasoning quantification CNN
CN112288744B (en) * 2020-08-24 2023-04-07 西安电子科技大学 SAR image change detection method based on integer reasoning quantification CNN
CN113033510A (en) * 2021-05-21 2021-06-25 浙江大华技术股份有限公司 Training and detecting method, device and storage medium for image change detection model
CN113569950A (en) * 2021-07-28 2021-10-29 大唐环境产业集团股份有限公司 Power station equipment fault monitoring model generation method, system and device
CN113569950B (en) * 2021-07-28 2024-05-28 大唐环境产业集团股份有限公司 Power station equipment fault monitoring model generation method, system and device
CN116206215A (en) * 2023-03-17 2023-06-02 银河航天(北京)网络技术有限公司 Forest land state monitoring method, forest land state monitoring device and storage medium
CN116206215B (en) * 2023-03-17 2023-09-29 银河航天(北京)网络技术有限公司 Forest land state monitoring method, forest land state monitoring device and storage medium
CN116990298A (en) * 2023-09-28 2023-11-03 南通中奥车用新材料有限公司 Finished product quality evaluation system for artificial leather production equipment
CN116990298B (en) * 2023-09-28 2023-12-08 南通中奥车用新材料有限公司 Finished product quality evaluation system for artificial leather production equipment

Also Published As

Publication number Publication date
CN108765465B (en) 2020-07-10

Similar Documents

Publication Publication Date Title
CN108765465A (en) A kind of unsupervised SAR image change detection
CN111209952B (en) Underwater target detection method based on improved SSD and migration learning
WO2019104767A1 (en) Fabric defect detection method based on deep convolutional neural network and visual saliency
De-Maeztu et al. Linear stereo matching
Muñoz et al. Strategies for image segmentation combining region and boundary information
Scharfenberger et al. Statistical textural distinctiveness for salient region detection in natural images
CN110119728A (en) Remote sensing images cloud detection method of optic based on Multiscale Fusion semantic segmentation network
CN109154978A (en) System and method for detecting plant disease
CN106780485A (en) SAR image change detection based on super-pixel segmentation and feature learning
CN107133943A (en) A kind of visible detection method of stockbridge damper defects detection
CN106875395B (en) Super-pixel-level SAR image change detection method based on deep neural network
CN108257151B (en) PCANet image change detection method based on significance analysis
CN109543688A (en) A kind of novel meter reading detection and knowledge method for distinguishing based on multilayer convolutional neural networks
CN109522908A (en) Image significance detection method based on area label fusion
Asokan et al. Machine learning based image processing techniques for satellite image analysis-a survey
CN108830832A (en) A kind of plastic barrel surface defects detection algorithm based on machine vision
CN107220649A (en) A kind of plain color cloth defects detection and sorting technique
CN106611423B (en) SAR image segmentation method based on ridge ripple filter and deconvolution structural model
TW200834459A (en) Video object segmentation method applied for rainy situations
CN108021869A (en) A kind of convolutional neural networks tracking of combination gaussian kernel function
CN109978848A (en) Method based on hard exudate in multiple light courcess color constancy model inspection eye fundus image
CN104616308A (en) Multiscale level set image segmenting method based on kernel fuzzy clustering
CN110728302A (en) Method for identifying color textile fabric tissue based on HSV (hue, saturation, value) and Lab (Lab) color spaces
CN107944403A (en) Pedestrian&#39;s attribute detection method and device in a kind of image
Chen et al. Single depth image super-resolution using convolutional neural networks

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant